Contingency severity assessment for voltage security using non-parametric regression techniques

This paper proposes a novel approach to power system voltage security assessment exploiting nonparametric regression techniques to extract simple, and at the same time reliable, models of the severity of a contingency, defined as the difference between pre- and post-contingency load power margins. The regression techniques extract information from large sets of possible operating conditions of a power system screened offline via massive random sampling, whose voltage security with respect to contingencies is pre-analyzed using an efficient voltage stability simulation. In particular, regression trees are used to identify the most salient parameters of the pre-contingency topology and electrical state which influence the severity of a given contingency, and to provide a first guess transparent approximation of the contingency severity in terms of these latter parameters. Multilayer perceptrons are exploited to further refine this information. The approach is demonstrated on a realistic model of a large scale voltage stability limited power system, where it shows to provide valuable physical insight and reliable contingency evaluation. Various potential uses in power system planning and operation are discussed